A crack characterization model for subsea pipeline based on spatial magnetic signals features

The inspection and quantification of minor defects for subsea pipelines is an important topic for ensuring structural stability and safety. This study proposes an in-line inspection and characterization model for the subsea pipeline cracks based on spatial magnetic signals extraction technology and machine learning algorithms. First, the crack inspection mechanism is elaborated by the magnetic domain deflection from the micro level, and the effect of the crack width, depth, and lift-off on the signal features are evaluated by numerical simulations. Second, a custom-built experimental system adopted to scan the tiny cracks is incorporated to validate the reliability of the simulation results. Finally, 11 statistic features are extracted from the time series signals and defined as the input of the regression models based on various machine learning algorithms to realize the quantitative characterization of the crack dimensions. Consequently, the probes manufactured in this research can accurately identify cracks with a minimum width of 0.3 mm. Moreover, the random forest algorithm achieves state-of-the-art inversion performance with mean errors of the prediction accuracy to the crack width, depth, and lift-off at 0.102 mm, 0.040 mm, and 0.004 mm, respectively.